rd-evaluate

name	rd-evaluate
description	Synthesize-stage: close iteration or continue back to framing in an existing standalone V3 run.
argument-hint	--run <run_id> --branch <branch_id> --recommendation <continue|stop> | --continue
allowed-tools	["Read","Bash"]

<codex_skill_adapter>

Invocation

• Claude Code: rd-evaluate skill
• Codex: $rd-evaluate
• Treat text after skill mention as arguments.

Tool mapping

• AskUserQuestion → request_user_input </codex_skill_adapter>

Complete or replay the synthesize stage for an existing V3 run and branch. Return the branch decision as continue (back to rd-propose) or stop. Never starts the full loop — use rd-agent for that.

Maps to rd_agent.entry.rd_evaluate.rd_evaluate.

<execution_context> @skills/rd-evaluate/workflows/continue.md @skills/_shared/references/stage-contract.md @skills/_shared/references/tool-execution-context.md </execution_context>

<required_fields>

• run_id (string): the run identifier for the paused standalone V3 run
• branch_id (string): the branch that owns the current synthesize step
• summary (string): the current-step summary to publish
• artifact_ids (list): the current-step artifact identifiers to publish or replay
• recommendation (string): "continue" or "stop" </required_fields>

Execute the continue workflow from @skills/rd-evaluate/workflows/continue.md end-to-end.

<outcome_guide>

• reused: published synthesize evidence still valid → honor existing recommendation
• review: synthesize needs manual review → surface reason, do NOT claim settled
• replay: synthesize needs fresh publish → replay then apply recommendation
• completed with continue: next skill is rd-propose (back to framing)
• completed with stop: loop ends, no next stage skill </outcome_guide>